Researchers assess 19 machine learning methods to identify and improve the optimal machine learning method for fall detection in the elderly in IoT-based intelligent environments, that achieves almost 100% accuracy.
CINCINNATI, OH / ACCESSWIRE / July 23, 2021 / Falls are a critical issue in the global aging population. Falls have severely affected the increasing aging population in the United States. According to the Centers for Disease Control and Prevention, about 36 million older adults fall each year, and every second of every day, an older adult (age 65+) suffers a fall in the U.S.-making falls the leading cause of injury and injury death in this age group. One out of four older adults will fall each year in the United States, making falls a public health concern, particularly among the aging population. Each year, in the United States, about $50 billion is spent on medical costs related to non-fatal fall injuries, and $754 million is spent related to fatal falls.
In addition to the health-related impacts, falls are associated with multiple behavioral, social, emotional, mental, and psychological impacts to the elderly such as - reduced mobility leading to loneliness and social isolation, fear of moving around, and loss of confidence in carrying out daily routine tasks in an independent manner. Thus, prompt, reliable, and accurate fall detection is very crucial to limit the severe and diverse consequences of falling both to the elderly and to various economies across the world in terms of the associated medical and caregiver costs.
While different machine learning approaches for fall detection have been investigated by researchers in the recent past, most of them have not been highly accurate, and none of these works focused on identifying the best machine learning approach. Moreover, most of the existing fall detection approaches, being binary classification systems tend to detect fall-like motions (for example, a posture where both hands and feet touch the ground) as falls, which raises false alarms in caregivers or medical personnel, which leads to them becoming desensitized to alarms for falls, causing a decrease in quality of care or even no timely care. In addition to this, the existing methodologies are not able to track a long lie after a fall. Lying on the floor for an extended period after falling can lead to serious complications; hence, detecting long lies is vital along with fall detection.
A duo of researchers from the University of Cincinnati, Nirmalya Thakur and Chia Y. Han, have proposed multiple novel methodologies and approaches for addressing all these challenges related to fall detection in the elderly in their recent paper titled - "A Study of Fall Detection in Assisted Living: Identifying and Improving the Optimal Machine Learning Method." "This machine learning-based fall detection system, which uses the k-folds cross-validation and the AdaBoost algorithm, achieves almost 100% overall accuracy and outperforms all related works in this field. This is the first time any work in this field has achieved almost 100% accuracy. Moreover, this is the first study in this area of research that involved the development, implementation, and evaluation of 19 different machine learning methods to determine the optimal machine learning method for fall detection. The work makes multiple other contributions to the area of assisted living as well, which, we hope, will ensure healthy aging of the elderly population in the future of Smart Homes," explains Thakur, the principal researcher of this study. Their groundbreaking study has been published in a recent issue of the Journal of Sensor and Actuator Networks.
At first, the researchers performed an extensive comparative study where they developed fall detection systems using all the 19 major machine learning algorithms - random forest, artificial neural network, decision tree, multiway decision tree, support vector machine, k-NN, gradient boosted trees, ID3, decision stump, CHAID, AutoMLP, linear regression, vector linear regression, random tree, naïve bayes, naïve bayes (kernel), linear discriminant analysis, quadratic discriminant analysis, and deep learning, and compared the respective overall accuracies and class precision values of these methods. Their study showed that the k-NN method is the optimal machine learning method for fall detection in terms of performance accuracy. Next, they used this k-NN method to develop a multi-label classifier to recognize postures and movements to distinctly identify falls and fall-like motions. This approach, therefore, resolves the issues with traditional binary classifiers. Further, the researchers used k-folds cross-validation with the AdaBoost algorithm to reduce false positives and overfitting of data while improving the performance accuracy of the k-NN-based multi-label classifier. When this approach was evaluated on two datasets, high-performance accuracies of 99.87% and 99.66% were achieved. Finally, the researchers concluded that the proposed approach could reliably identify the action of standing up from a lying position, which, they are confident, will help in determining whether a fall was accompanied by a long lie.
As the world is rapidly moving towards the new era of interconnected homes and cities through the Internet of Things and Artificial Intelligence, a fully functional smart home is closer to reality than ever before. Thakur is optimistic that their work will offer a plethora of elderly care opportunities in such IoT-based intelligent living environments, "Owing to the high accuracy, reliability, and robustness of our fall detection system, we expect that it can be seamlessly integrated into a myriad of wearable devices and smart gadgets to track, study, analyze, and detect falls and multiple other health-related needs of the elderly, to contribute towards their assisted living and healthy aging," he comments.
Figure: Summary of the methodologies and the multiple scientific contributions of this research work. Figure courtesy: Nirmalya Thakur from the University of Cincinnati
The scientific contributions of this study could be of paramount importance in terms of intelligent assisted living in the future of Smart Homes and Smart Cities. We hope this paradigm shift will take place soon. After all, none of us are getting any younger!
Authors: Nirmalya Thakur and Chia Y. Han
Title of the original paper: A Study of Fall Detection in Assisted Living: Identifying and Improving the Optimal Machine Learning Method
Journal: Journal of Sensor and Actuator Networks
Affiliations: Department of Electrical Engineering and Computer Science, University of Cincinnati, Ohio, USA
About Nirmalya Thakur
Nirmalya Thakur, an internationally acclaimed expert in the area of Human-Computer Interaction and its interrelated disciplines, is currently working as an instructor at the Department of Electrical Engineering and Computer Science at the University of Cincinnati in Ohio, United States. Aside from Human-Computer Interaction, his research interests include Artificial Intelligence, Machine Learning, Internet of Things, Data Science, and Natural Language Processing. He has authored 31 peer-reviewed publications in these fields and has reviewed over 320 papers by serving as a reviewer for different conferences and journals. Thakur has won several national and international awards, including the Young Scientist of the Year Award, the Young Innovator Award, the Excellent Researcher Award, and the Research Excellence Award. Last year he was inducted into the elite list of Marquis Top Scientists of the World for his exemplary contributions to the field of Computer Science.
SOURCE: Nirmalya Thakur